Comparison of Molecular Recognition in Docking Versus Experimental CSD and PDB Data

Molecular docking is a widely used technique in structure-based drug design for generating poses of small molecules in a protein receptor structure. These poses are then ranked to prioritize compounds for experimental validation. Numerous approaches to assessing the structural fit of a ligand exist, ranging from simple scoring functions to more elaborate free energy calculations. Regardless of the prioritization method chosen, its accuracy is limited by the quality of the protein-ligand pose. Here, we apply two established statistical approaches for quantifying atomic interaction preferences and torsional ligand strain, respectively, to compare poses generated by the docking algorithm Vina with crystallographic data from the PDB and CSD. This analysis allows us to identify potential deficiencies in the docking algorithm, such as underestimated electrostatic repulsion or high-energy hydroxyl conformations. By highlighting such inaccuracies, we aim to inspire improvements in future docking algorithms. Finally, a pose scoring approach is proposed that significantly improves the retrieval of the experimental pose from a set of docked poses.